Show More
@@ -1,198 +1,232 b'' | |||||
1 | # setdiscovery.py - improved discovery of common nodeset for mercurial |
|
1 | # setdiscovery.py - improved discovery of common nodeset for mercurial | |
2 | # |
|
2 | # | |
3 | # Copyright 2010 Benoit Boissinot <bboissin@gmail.com> |
|
3 | # Copyright 2010 Benoit Boissinot <bboissin@gmail.com> | |
4 | # and Peter Arrenbrecht <peter@arrenbrecht.ch> |
|
4 | # and Peter Arrenbrecht <peter@arrenbrecht.ch> | |
5 | # |
|
5 | # | |
6 | # This software may be used and distributed according to the terms of the |
|
6 | # This software may be used and distributed according to the terms of the | |
7 | # GNU General Public License version 2 or any later version. |
|
7 | # GNU General Public License version 2 or any later version. | |
|
8 | """ | |||
|
9 | Algorithm works in the following way. You have two repository: local and | |||
|
10 | remote. They both contains a DAG of changelists. | |||
|
11 | ||||
|
12 | The goal of the discovery protocol is to find one set of node *common*, | |||
|
13 | the set of nodes shared by local and remote. | |||
|
14 | ||||
|
15 | One of the issue with the original protocol was latency, it could | |||
|
16 | potentially require lots of roundtrips to discover that the local repo was a | |||
|
17 | subset of remote (which is a very common case, you usually have few changes | |||
|
18 | compared to upstream, while upstream probably had lots of development). | |||
|
19 | ||||
|
20 | The new protocol only requires one interface for the remote repo: `known()`, | |||
|
21 | which given a set of changelists tells you if they are present in the DAG. | |||
|
22 | ||||
|
23 | The algorithm then works as follow: | |||
|
24 | ||||
|
25 | - We will be using three sets, `common`, `missing`, `unknown`. Originally | |||
|
26 | all nodes are in `unknown`. | |||
|
27 | - Take a sample from `unknown`, call `remote.known(sample)` | |||
|
28 | - For each node that remote knows, move it and all its ancestors to `common` | |||
|
29 | - For each node that remote doesn't know, move it and all its descendants | |||
|
30 | to `missing` | |||
|
31 | - Iterate until `unknown` is empty | |||
|
32 | ||||
|
33 | There are a couple optimizations, first is instead of starting with a random | |||
|
34 | sample of missing, start by sending all heads, in the case where the local | |||
|
35 | repo is a subset, you computed the answer in one round trip. | |||
|
36 | ||||
|
37 | Then you can do something similar to the bisecting strategy used when | |||
|
38 | finding faulty changesets. Instead of random samples, you can try picking | |||
|
39 | nodes that will maximize the number of nodes that will be | |||
|
40 | classified with it (since all ancestors or descendants will be marked as well). | |||
|
41 | """ | |||
8 |
|
42 | |||
9 | from node import nullid |
|
43 | from node import nullid | |
10 | from i18n import _ |
|
44 | from i18n import _ | |
11 | import random |
|
45 | import random | |
12 | import util, dagutil |
|
46 | import util, dagutil | |
13 |
|
47 | |||
14 | def _updatesample(dag, nodes, sample, always, quicksamplesize=0): |
|
48 | def _updatesample(dag, nodes, sample, always, quicksamplesize=0): | |
15 | # if nodes is empty we scan the entire graph |
|
49 | # if nodes is empty we scan the entire graph | |
16 | if nodes: |
|
50 | if nodes: | |
17 | heads = dag.headsetofconnecteds(nodes) |
|
51 | heads = dag.headsetofconnecteds(nodes) | |
18 | else: |
|
52 | else: | |
19 | heads = dag.heads() |
|
53 | heads = dag.heads() | |
20 | dist = {} |
|
54 | dist = {} | |
21 | visit = util.deque(heads) |
|
55 | visit = util.deque(heads) | |
22 | seen = set() |
|
56 | seen = set() | |
23 | factor = 1 |
|
57 | factor = 1 | |
24 | while visit: |
|
58 | while visit: | |
25 | curr = visit.popleft() |
|
59 | curr = visit.popleft() | |
26 | if curr in seen: |
|
60 | if curr in seen: | |
27 | continue |
|
61 | continue | |
28 | d = dist.setdefault(curr, 1) |
|
62 | d = dist.setdefault(curr, 1) | |
29 | if d > factor: |
|
63 | if d > factor: | |
30 | factor *= 2 |
|
64 | factor *= 2 | |
31 | if d == factor: |
|
65 | if d == factor: | |
32 | if curr not in always: # need this check for the early exit below |
|
66 | if curr not in always: # need this check for the early exit below | |
33 | sample.add(curr) |
|
67 | sample.add(curr) | |
34 | if quicksamplesize and (len(sample) >= quicksamplesize): |
|
68 | if quicksamplesize and (len(sample) >= quicksamplesize): | |
35 | return |
|
69 | return | |
36 | seen.add(curr) |
|
70 | seen.add(curr) | |
37 | for p in dag.parents(curr): |
|
71 | for p in dag.parents(curr): | |
38 | if not nodes or p in nodes: |
|
72 | if not nodes or p in nodes: | |
39 | dist.setdefault(p, d + 1) |
|
73 | dist.setdefault(p, d + 1) | |
40 | visit.append(p) |
|
74 | visit.append(p) | |
41 |
|
75 | |||
42 | def _setupsample(dag, nodes, size): |
|
76 | def _setupsample(dag, nodes, size): | |
43 | if len(nodes) <= size: |
|
77 | if len(nodes) <= size: | |
44 | return set(nodes), None, 0 |
|
78 | return set(nodes), None, 0 | |
45 | always = dag.headsetofconnecteds(nodes) |
|
79 | always = dag.headsetofconnecteds(nodes) | |
46 | desiredlen = size - len(always) |
|
80 | desiredlen = size - len(always) | |
47 | if desiredlen <= 0: |
|
81 | if desiredlen <= 0: | |
48 | # This could be bad if there are very many heads, all unknown to the |
|
82 | # This could be bad if there are very many heads, all unknown to the | |
49 | # server. We're counting on long request support here. |
|
83 | # server. We're counting on long request support here. | |
50 | return always, None, desiredlen |
|
84 | return always, None, desiredlen | |
51 | return always, set(), desiredlen |
|
85 | return always, set(), desiredlen | |
52 |
|
86 | |||
53 | def _takequicksample(dag, nodes, size, initial): |
|
87 | def _takequicksample(dag, nodes, size, initial): | |
54 | always, sample, desiredlen = _setupsample(dag, nodes, size) |
|
88 | always, sample, desiredlen = _setupsample(dag, nodes, size) | |
55 | if sample is None: |
|
89 | if sample is None: | |
56 | return always |
|
90 | return always | |
57 | if initial: |
|
91 | if initial: | |
58 | fromset = None |
|
92 | fromset = None | |
59 | else: |
|
93 | else: | |
60 | fromset = nodes |
|
94 | fromset = nodes | |
61 | _updatesample(dag, fromset, sample, always, quicksamplesize=desiredlen) |
|
95 | _updatesample(dag, fromset, sample, always, quicksamplesize=desiredlen) | |
62 | sample.update(always) |
|
96 | sample.update(always) | |
63 | return sample |
|
97 | return sample | |
64 |
|
98 | |||
65 | def _takefullsample(dag, nodes, size): |
|
99 | def _takefullsample(dag, nodes, size): | |
66 | always, sample, desiredlen = _setupsample(dag, nodes, size) |
|
100 | always, sample, desiredlen = _setupsample(dag, nodes, size) | |
67 | if sample is None: |
|
101 | if sample is None: | |
68 | return always |
|
102 | return always | |
69 | # update from heads |
|
103 | # update from heads | |
70 | _updatesample(dag, nodes, sample, always) |
|
104 | _updatesample(dag, nodes, sample, always) | |
71 | # update from roots |
|
105 | # update from roots | |
72 | _updatesample(dag.inverse(), nodes, sample, always) |
|
106 | _updatesample(dag.inverse(), nodes, sample, always) | |
73 | assert sample |
|
107 | assert sample | |
74 | if len(sample) > desiredlen: |
|
108 | if len(sample) > desiredlen: | |
75 | sample = set(random.sample(sample, desiredlen)) |
|
109 | sample = set(random.sample(sample, desiredlen)) | |
76 | elif len(sample) < desiredlen: |
|
110 | elif len(sample) < desiredlen: | |
77 | more = desiredlen - len(sample) |
|
111 | more = desiredlen - len(sample) | |
78 | sample.update(random.sample(list(nodes - sample - always), more)) |
|
112 | sample.update(random.sample(list(nodes - sample - always), more)) | |
79 | sample.update(always) |
|
113 | sample.update(always) | |
80 | return sample |
|
114 | return sample | |
81 |
|
115 | |||
82 | def findcommonheads(ui, local, remote, |
|
116 | def findcommonheads(ui, local, remote, | |
83 | initialsamplesize=100, |
|
117 | initialsamplesize=100, | |
84 | fullsamplesize=200, |
|
118 | fullsamplesize=200, | |
85 | abortwhenunrelated=True): |
|
119 | abortwhenunrelated=True): | |
86 | '''Return a tuple (common, anyincoming, remoteheads) used to identify |
|
120 | '''Return a tuple (common, anyincoming, remoteheads) used to identify | |
87 | missing nodes from or in remote. |
|
121 | missing nodes from or in remote. | |
88 | ''' |
|
122 | ''' | |
89 | roundtrips = 0 |
|
123 | roundtrips = 0 | |
90 | cl = local.changelog |
|
124 | cl = local.changelog | |
91 | dag = dagutil.revlogdag(cl) |
|
125 | dag = dagutil.revlogdag(cl) | |
92 |
|
126 | |||
93 | # early exit if we know all the specified remote heads already |
|
127 | # early exit if we know all the specified remote heads already | |
94 | ui.debug("query 1; heads\n") |
|
128 | ui.debug("query 1; heads\n") | |
95 | roundtrips += 1 |
|
129 | roundtrips += 1 | |
96 | ownheads = dag.heads() |
|
130 | ownheads = dag.heads() | |
97 | sample = ownheads |
|
131 | sample = ownheads | |
98 | if remote.local(): |
|
132 | if remote.local(): | |
99 | # stopgap until we have a proper localpeer that supports batch() |
|
133 | # stopgap until we have a proper localpeer that supports batch() | |
100 | srvheadhashes = remote.heads() |
|
134 | srvheadhashes = remote.heads() | |
101 | yesno = remote.known(dag.externalizeall(sample)) |
|
135 | yesno = remote.known(dag.externalizeall(sample)) | |
102 | elif remote.capable('batch'): |
|
136 | elif remote.capable('batch'): | |
103 | batch = remote.batch() |
|
137 | batch = remote.batch() | |
104 | srvheadhashesref = batch.heads() |
|
138 | srvheadhashesref = batch.heads() | |
105 | yesnoref = batch.known(dag.externalizeall(sample)) |
|
139 | yesnoref = batch.known(dag.externalizeall(sample)) | |
106 | batch.submit() |
|
140 | batch.submit() | |
107 | srvheadhashes = srvheadhashesref.value |
|
141 | srvheadhashes = srvheadhashesref.value | |
108 | yesno = yesnoref.value |
|
142 | yesno = yesnoref.value | |
109 | else: |
|
143 | else: | |
110 | # compatibility with pre-batch, but post-known remotes during 1.9 |
|
144 | # compatibility with pre-batch, but post-known remotes during 1.9 | |
111 | # development |
|
145 | # development | |
112 | srvheadhashes = remote.heads() |
|
146 | srvheadhashes = remote.heads() | |
113 | sample = [] |
|
147 | sample = [] | |
114 |
|
148 | |||
115 | if cl.tip() == nullid: |
|
149 | if cl.tip() == nullid: | |
116 | if srvheadhashes != [nullid]: |
|
150 | if srvheadhashes != [nullid]: | |
117 | return [nullid], True, srvheadhashes |
|
151 | return [nullid], True, srvheadhashes | |
118 | return [nullid], False, [] |
|
152 | return [nullid], False, [] | |
119 |
|
153 | |||
120 | # start actual discovery (we note this before the next "if" for |
|
154 | # start actual discovery (we note this before the next "if" for | |
121 | # compatibility reasons) |
|
155 | # compatibility reasons) | |
122 | ui.status(_("searching for changes\n")) |
|
156 | ui.status(_("searching for changes\n")) | |
123 |
|
157 | |||
124 | srvheads = dag.internalizeall(srvheadhashes, filterunknown=True) |
|
158 | srvheads = dag.internalizeall(srvheadhashes, filterunknown=True) | |
125 | if len(srvheads) == len(srvheadhashes): |
|
159 | if len(srvheads) == len(srvheadhashes): | |
126 | ui.debug("all remote heads known locally\n") |
|
160 | ui.debug("all remote heads known locally\n") | |
127 | return (srvheadhashes, False, srvheadhashes,) |
|
161 | return (srvheadhashes, False, srvheadhashes,) | |
128 |
|
162 | |||
129 | if sample and util.all(yesno): |
|
163 | if sample and util.all(yesno): | |
130 | ui.note(_("all local heads known remotely\n")) |
|
164 | ui.note(_("all local heads known remotely\n")) | |
131 | ownheadhashes = dag.externalizeall(ownheads) |
|
165 | ownheadhashes = dag.externalizeall(ownheads) | |
132 | return (ownheadhashes, True, srvheadhashes,) |
|
166 | return (ownheadhashes, True, srvheadhashes,) | |
133 |
|
167 | |||
134 | # full blown discovery |
|
168 | # full blown discovery | |
135 |
|
169 | |||
136 | # own nodes where I don't know if remote knows them |
|
170 | # own nodes where I don't know if remote knows them | |
137 | undecided = dag.nodeset() |
|
171 | undecided = dag.nodeset() | |
138 | # own nodes I know we both know |
|
172 | # own nodes I know we both know | |
139 | common = set() |
|
173 | common = set() | |
140 | # own nodes I know remote lacks |
|
174 | # own nodes I know remote lacks | |
141 | missing = set() |
|
175 | missing = set() | |
142 |
|
176 | |||
143 | # treat remote heads (and maybe own heads) as a first implicit sample |
|
177 | # treat remote heads (and maybe own heads) as a first implicit sample | |
144 | # response |
|
178 | # response | |
145 | common.update(dag.ancestorset(srvheads)) |
|
179 | common.update(dag.ancestorset(srvheads)) | |
146 | undecided.difference_update(common) |
|
180 | undecided.difference_update(common) | |
147 |
|
181 | |||
148 | full = False |
|
182 | full = False | |
149 | while undecided: |
|
183 | while undecided: | |
150 |
|
184 | |||
151 | if sample: |
|
185 | if sample: | |
152 | commoninsample = set(n for i, n in enumerate(sample) if yesno[i]) |
|
186 | commoninsample = set(n for i, n in enumerate(sample) if yesno[i]) | |
153 | common.update(dag.ancestorset(commoninsample, common)) |
|
187 | common.update(dag.ancestorset(commoninsample, common)) | |
154 |
|
188 | |||
155 | missinginsample = [n for i, n in enumerate(sample) if not yesno[i]] |
|
189 | missinginsample = [n for i, n in enumerate(sample) if not yesno[i]] | |
156 | missing.update(dag.descendantset(missinginsample, missing)) |
|
190 | missing.update(dag.descendantset(missinginsample, missing)) | |
157 |
|
191 | |||
158 | undecided.difference_update(missing) |
|
192 | undecided.difference_update(missing) | |
159 | undecided.difference_update(common) |
|
193 | undecided.difference_update(common) | |
160 |
|
194 | |||
161 | if not undecided: |
|
195 | if not undecided: | |
162 | break |
|
196 | break | |
163 |
|
197 | |||
164 | if full: |
|
198 | if full: | |
165 | ui.note(_("sampling from both directions\n")) |
|
199 | ui.note(_("sampling from both directions\n")) | |
166 | sample = _takefullsample(dag, undecided, size=fullsamplesize) |
|
200 | sample = _takefullsample(dag, undecided, size=fullsamplesize) | |
167 | elif common: |
|
201 | elif common: | |
168 | # use cheapish initial sample |
|
202 | # use cheapish initial sample | |
169 | ui.debug("taking initial sample\n") |
|
203 | ui.debug("taking initial sample\n") | |
170 | sample = _takefullsample(dag, undecided, size=fullsamplesize) |
|
204 | sample = _takefullsample(dag, undecided, size=fullsamplesize) | |
171 | else: |
|
205 | else: | |
172 | # use even cheaper initial sample |
|
206 | # use even cheaper initial sample | |
173 | ui.debug("taking quick initial sample\n") |
|
207 | ui.debug("taking quick initial sample\n") | |
174 | sample = _takequicksample(dag, undecided, size=initialsamplesize, |
|
208 | sample = _takequicksample(dag, undecided, size=initialsamplesize, | |
175 | initial=True) |
|
209 | initial=True) | |
176 |
|
210 | |||
177 | roundtrips += 1 |
|
211 | roundtrips += 1 | |
178 | ui.progress(_('searching'), roundtrips, unit=_('queries')) |
|
212 | ui.progress(_('searching'), roundtrips, unit=_('queries')) | |
179 | ui.debug("query %i; still undecided: %i, sample size is: %i\n" |
|
213 | ui.debug("query %i; still undecided: %i, sample size is: %i\n" | |
180 | % (roundtrips, len(undecided), len(sample))) |
|
214 | % (roundtrips, len(undecided), len(sample))) | |
181 | # indices between sample and externalized version must match |
|
215 | # indices between sample and externalized version must match | |
182 | sample = list(sample) |
|
216 | sample = list(sample) | |
183 | yesno = remote.known(dag.externalizeall(sample)) |
|
217 | yesno = remote.known(dag.externalizeall(sample)) | |
184 | full = True |
|
218 | full = True | |
185 |
|
219 | |||
186 | result = dag.headsetofconnecteds(common) |
|
220 | result = dag.headsetofconnecteds(common) | |
187 | ui.progress(_('searching'), None) |
|
221 | ui.progress(_('searching'), None) | |
188 | ui.debug("%d total queries\n" % roundtrips) |
|
222 | ui.debug("%d total queries\n" % roundtrips) | |
189 |
|
223 | |||
190 | if not result and srvheadhashes != [nullid]: |
|
224 | if not result and srvheadhashes != [nullid]: | |
191 | if abortwhenunrelated: |
|
225 | if abortwhenunrelated: | |
192 | raise util.Abort(_("repository is unrelated")) |
|
226 | raise util.Abort(_("repository is unrelated")) | |
193 | else: |
|
227 | else: | |
194 | ui.warn(_("warning: repository is unrelated\n")) |
|
228 | ui.warn(_("warning: repository is unrelated\n")) | |
195 | return (set([nullid]), True, srvheadhashes,) |
|
229 | return (set([nullid]), True, srvheadhashes,) | |
196 |
|
230 | |||
197 | anyincoming = (srvheadhashes != [nullid]) |
|
231 | anyincoming = (srvheadhashes != [nullid]) | |
198 | return dag.externalizeall(result), anyincoming, srvheadhashes |
|
232 | return dag.externalizeall(result), anyincoming, srvheadhashes |
General Comments 0
You need to be logged in to leave comments.
Login now